Summary
To build a free AI chatbot for support, train it on your help center and docs, set how it answers, give it a handover to a person and add it to your site; Wonderchat's free plan covers all four and refills 50 AI replies every month. How many questions it resolves depends less on the AI model than on how complete your knowledge base is and how well it hands over to a person.
The quality of your knowledge base is the single most important factor for success. Treat your documentation like a core product to avoid shallow answers and AI hallucinations.
Prevent customer frustration by building a smart "human handover" safety net. When the AI can't resolve an issue, it must escalate to a human expert with full context.
Wonderchat resolves about 75% of evaluated conversations without a person; Jortt, a Dutch accounting platform, resolves about 92% of roughly 30,000 monthly inquiries without a person, and Encompass runs it alongside Zendesk.
Your inbox is full of the same questions over and over, and the answers are already somewhere in your docs. Your support inbox is full of questions from people who can't find the right document, pricing page, or contact form. Someone convinced you to "just set up a chatbot."
So you did. And now customers are getting routed to a generic FAQ page, or worse, receiving confidently wrong answers because the bot hallucinated its way through a gap in your docs.
Sound familiar? You're not alone. The problem isn't AI. It's that most free AI chatbot setups are optimized to keep tickets out of the queue, not to resolve them.
There's a critical difference. Keeping a ticket out of the queue is when the bot technically responds but doesn't actually solve anything. Resolution is when the user finds what they need, whether it's a support answer, a product spec, or the right person to talk to.
The goal isn't to build another set of glorified FAQs. It's to build an agent that answers from your whole knowledge base, says so when your docs don't cover a question, and hands the person to your team when they need one.
That requires two things working together: a deep, comprehensive knowledge base and a smooth human fallback. Get both right and you have the setup Wonderchat customers like Jortt, Ko-fi and Encompass run: answers from their own docs, with a person one handover away.
Here's the exact blueprint.
The 4-Step Blueprint
Step 1: Feed Your AI a Comprehensive Knowledge Base
The most common reason AI chatbots underperform has nothing to do with the AI model: it's what you feed it. Most teams spend their energy picking the tool and very little on the knowledge base, when it should be the other way round. A shallow knowledge base produces shallow answers.
Here's how to train it on real business knowledge:
In your Wonderchat dashboard, click "Create Agent" on the chatbots page.
Give your chatbot a Name, Tagline, and brand Color.
Under Data Sources, choose how you want to train it:
Website Links
Paste your website URL and Wonderchat crawls it automatically. It supports full-site crawls, specific pages and blog sections, and on the Scale plan it re-crawls weekly so your agent stays current as your content changes.
File Uploads
Upload your support docs in the following formats:
PDF
DOCX
TXT
CSV
PPT
Wonderchat is built to handle large knowledge bases, from technical documentation to product catalogs and policy manuals. If you have complex content, this matters.
Your Help Center
Point Wonderchat at your help center, on Zendesk or anywhere else, to pull in the articles you already have. This gives your AI an instant head start with knowledge your team has already curated.
The result is an agent that answers from your own content rather than generic training data, and shows the page each answer came from.
Step 2: Configure Your AI's Persona for On-Brand Conversations
An AI that gives correct answers in a robotic, off-brand tone still creates a bad customer experience. This step makes your agent feel like a natural extension of your team.
From the Wonderchat setup guide:
Select Your AI Model: the model is a setting, by plan, with models from OpenAI
Anthropic
Google
DeepSeek. Answers still come only from your content, and you can switch models as your needs change.
Define the Agent's Role: Write a clear instruction prompt. For example: "You are a friendly support agent for [Company]. Answer questions concisely and accurately using the provided knowledge base. If you're unsure, say so and offer to escalate."
Set Default Language: Wonderchat answers in 40+ languages, useful if you serve customers in more than one.
Craft Greeting and Fallback Messages: Your greeting sets the tone. Your fallback message, what the bot says when it can't find an answer, is equally important. A good fallback doesn't leave the user stranded; it guides them toward a human. Which brings the guide to the most important step.
Step 3: Build a Safety Net with Smart Human Handover
Over-automation is a real failure mode. When an AI can't help and offers no escape route, customers don't just give up, they get angry, leave, and re-open tickets through other channels.
The fix isn't less AI. It's smarter routing. A system that knows when and how to hand off to a human expert.
Wonderchat is built for this: the agent answers, and a person is one handover away. On every plan the agent hands over by ticket or email with the full chat attached, and from Basic, live chat lets a teammate take over in real time inside the same product.
Here's how to configure it in Wonderchat:
Go to Chatbots > Actions (⋮) > Edit Chatbot.
Toggle on "Enable Human Handover", this adds a live escalation option directly to your chat widget.
Set Automated Triggers: After a set number of messages (for example, 3 to 5), a feedback prompt appears automatically.
When no page covers the question, the agent says so instead of guessing and offers a person, so no customer is left in a loop.
Configure Handover Channels:
Email: Route escalations directly to your support team's inbox.
Zendesk Integration: Automatically create tickets with full conversation context attached. Encompass runs Wonderchat as a direct extension of its Zendesk help desk: the agent handles Tier-1 questions and routes complex issues to Zendesk with everything the human agent needs to pick up immediately.
Built-in Live Chat (Basic and up): a teammate takes over the conversation in real time inside Wonderchat, with no chat history copied between tools.
Customize Your Handover Form: Add fields such as Name or Issue Type that your team needs before picking up the conversation.
The outcome: users who can self-serve find what they need instantly. The ones who need an expert get routed to the right person, with full context, without friction.
Step 4: Deploy the Same Day
Once configured, click "Create". Wonderchat generates an embed script you can drop into any website, WordPress, Shopify, Squarespace, or raw HTML. 80% of Wonderchat agents that go live do so the same day, though knowledge bases over 500 pages take a median 15 days to go live.
Your AI agent is now live.
Why Most Free AI Chatbots Underperform (And How Yours Won't)
Knowing how these setups fail is what separates a bot that frustrates customers from one that answers them.
Pitfall #1: Shallow Knowledge Bases Lead to Confidently Wrong Answers
This is the most common and most damaging problem. When your knowledge base only covers surface-level FAQs, the AI will either fail to answer or, worse, hallucinate a plausible-sounding but incorrect response. As one support professional put it on Reddit: "Bad or incomplete documentation means even a great AI will confidently give customers the wrong answer, treat your knowledge base like the actual product."
RAG-based agents like Wonderchat retrieve answers from your uploaded content instead of generating them from thin air. But they're only as good as what you give them.
Treat your knowledge base like your actual product: keep it updated, structured, and comprehensive. Wonderchat answers from your documents and shows the source on every reply, so a user with a niche technical question can open the page the answer came from.
Pitfall #2: No Live Fallback Creates an "AI Dead End"
Simple chatbot tools are AI-only. When they can't find an answer in the knowledge base, they have no other routes to offer. That's the dead end, and customers who hit it don't quietly accept it. They churn, they escalate through email, or they leave a bad review.
Effective AI support is a hybrid model: the AI automates resolution for repetitive Tier-1 tickets, while humans handle the complex, high-stakes conversations that actually require judgment. With Wonderchat, that handover happens inside one product: by ticket or email with the full chat attached on every plan, or by live chat from Basic. The question that hit the dead end also lands in Wonderchat's review inbox, so you answer it once and the agent answers it from then on.
Pitfall #3: Poor Handling of Complex Queries and Varied Phrasing
Older chatbots relied on keyword matching. A customer who asks "how do I cancel" gets help. A customer who asks "I want to stop my subscription" gets a blank stare.
A good agent understands what the customer means, not just the words they used. They use that intent to route users to the right answer, in the right format. Wonderchat handles the different ways customers phrase the same question and answers from the matching page in your docs, with the source shown so they can open it.
Realistic Expectations: Your First Three Months
Setting honest expectations matters. Here's what a typical rollout looks like, from the first week to the third month.
Week 1: The Foundation
Your AI is live and handling its first real conversations. At this stage, it will answer the questions your knowledge base covers well and expose the gaps you didn't know existed. Open Wonderchat's review inbox to see the questions it couldn't answer, and answer each one once: paste a URL, pick a file or write it. The learning loop starts here: every failed response is a signal about what to add next.
Don't optimize for perfection at this stage. Optimize for coverage and handover reliability.
Month 1: The Refinement
You've uploaded the documents that filled your Week 1 gaps. Your handover triggers are tuned based on real conversation data. The AI is now confidently handling the majority of repetitive tickets, password resets, billing FAQs, shipping status questions, policy lookups, while routing the genuinely complex queries to your team with full context.
Metrics to watch: First Contact Resolution (FCR) and Time to Resolution (TTR). Both should start improving meaningfully. This is the phase where you can show your founder what it answers, what it can't, and what it costs.
Month 3: A Mature Agent
Your AI agent is now mature. It answers what your docs cover and hands over the rest. Your human team is no longer answering the same question for the 400th time, they're focused on the complex, high-value conversations that actually require expertise. Customers get answers from your docs at any hour, without waiting in a queue.
Two Wonderchat customers show what a mature agent looks like:
Jortt, a Dutch accounting platform: its agent "Femke" resolves about 92% of roughly 30,000 monthly inquiries without a person. Their support team's words: the remaining 8% is "far more interesting work."
Encompass (technical services): Resolves 75% of tickets from their Zendesk queue without a person, running Wonderchat as a direct AI extension of their helpdesk.
Across these deployments, the goal is a successful outcome in one conversation. Not a link to another page. Not a ticket opened and left pending. One conversation, one right answer, one completed user journey.
Your Agent Can Start Today
Building a support agent that actually works isn't about finding a magic tool and flipping a switch. It's a process:
Start with a deep, well-maintained knowledge base.
Configure a persona that reflects your brand.
Build intelligent routing to human experts so no customer ever hits a dead end.
Answer the questions that land in the review inbox, once each.
Done right, the agent stops being a simple Q&A bot: it answers what your docs cover at any hour, tells you what they don't, and hands the rest to your team.
Wonderchat's free plan refills 50 AI replies every month, so you can get your first agent live the same day. No credit card. No complex setup. Just connect your knowledge source, configure your agent, and deploy.
If Jortt, a Dutch accounting platform, resolves about 92% of roughly 30,000 monthly inquiries without a person, getting your users to the right answer is a very solvable problem.
Frequently Asked Questions
What is the difference between a bot that replies and one that resolves?
A bot that only replies responds without solving the problem; a bot that resolves gives the person the correct answer they were looking for. This blueprint aims for resolution, so people find what they need instead of a generic page.
How much content do I need to train an effective AI chatbot?
The more complete your documentation, the better the agent performs. Start with your existing FAQs and key support articles, then add what the review inbox shows is missing; Wonderchat handles large knowledge bases and niche questions.
How does Wonderchat prevent the AI from giving wrong answers or "hallucinating"?
Wonderchat uses retrieval-augmented generation (RAG): it answers only from the content you provide, not from generic training data. Every reply shows its source, and when no page covers a question, it says so instead of guessing.
What happens when the AI chatbot cannot answer a user's question?
When an AI chatbot cannot find an answer, it should escalate the conversation to a human. Wonderchat hands the conversation to your team with the full chat attached: by email, as a ticket in Wonderchat or Zendesk, or by live chat from Basic.
How do I keep the AI's knowledge base up to date?
On Wonderchat's Scale plan, your website and help center are re-crawled weekly, so new or changed pages reach the agent without manual work. On any plan, when a question lands in the review inbox, you answer it once by pasting a URL, picking a file or writing the answer.
Why is a hybrid AI and live chat system better than an AI-only chatbot?
An AI-only chatbot leaves people with complex issues at a dead end. With a handover, the agent answers the repeat questions at any hour and passes anyone who needs a person to your team with the conversation attached.
